Reconciling Translational Invariance and Hierarchy

Abstract: Tensor networks are not only numerical tools for describing ground states of quantum many-body systems, but also conceptual aids for understanding their entanglement structures. The proper way to understand tensor networks themselves is through explicit examples of solvable ground states that they describe exactly. In fact, this has historically been how tensor networks for gapped ground states, such as the matrix product state (MPS) and the projected entangled paired state, emerged as an elegant analytical framework from numerical techniques like the density matrix renormalization group. However, for gapless ground states, generically described by the multiscale entanglement renormalization ansatz (MERA), a corresponding exactly solvable model has so far been missing. This is because the hierarchical structure of MERA intrinsically breaks the translational invariance. We identify a condition for MERA to be compatible with translational invariance by examining equivalent networks of rank-3 tensors. The condition is satisfied by the previously constructed hierarchical tensor network for the Motzkin and Fredkin chains, which can be considered a non-unitary generalization to the MERA. The hierarchical TN description is complemented by a translationally invariant MPS alternative, which is used to derive the power-law decay of the correlation function and critical exponents.